At the same time, fashion is an industry where artificial intelligence is faced with several levels of complexity: interaction with the real, physical world; automation of processes over thousands of years based on manual work; aesthetic preferences that are poorly analysed by machine; and, finally, human rights issues to their own image. That’s why talking about AI in fashion can’t be reduced to a simple «he will replace designers» or «this is another smarter computer.» In this material in two parts, we’re going to try to sort out what’s going on, where AI has already become a working infrastructure, where it remains an experiment, and where it’s the source of new conflicts and questions. Based on academic research and evidence from existing projects, we will describe how the fashion industry uses artificial intelligence, ranging from design and production to filming, digital models and marketing language.

The word «artificial intelligence» in fashion refers to several different technologies: Machine learning, an algorithm that learns from data and projections, for example, changes in seasonal demand, sales volumes, probability of returns. Computer vision is a system that analyzes images and videos: tissue quality control, impairment recognition, silhouette analysis, virtual fitting. General AI are models that create new content: images, texts, videos, 3D objects. They are the most visible today. Large language models (LLM) are systems that work with text: product descriptions, marketing texts, brand platforms, internal instructions, control of legal requirements. Fashion researchers and the textile industry emphasize that it is not right to talk about «AI’s influence on fashion» as a single process. Each of these technologies is embedded in different segments of the industry chain and has different effects, ranging from saving resources to influencing the perception of fashion in general [1].

General AI as «first draft»

Together, these services show how the generation AI is gradually starting to act as a link between visual search and product preparation. Researchers emphasize that the greatest value of AI is acquired when it is linked to the design systems of the healer (CAD systems) to generate sketches with the required system of parameters [3].

Design is one of the most accurate processes of the industry chain, since a millimeter error can lead to a spoiled batch of products and kilometres of fabric. Automation of the cure therefore remains an area where AI is being used very carefully. Algorithms are good at basic forms and tailoring to individual body parameters. This is particularly required in mass caste production [4]. In practice, mass casteization is already being used by major industry players. For example, Leviøs adapts basic jean models to individual length and landing parameters, developing the idea of «ideal denim» within the mass market. Hugo Boss integrates a made-to-order and a made-to-measure into factory processes, through digital leashes and automation without mediating the hotel. Nike offers casteization of color and details, making the buyer a co-author. This model is increasingly being implemented not directly by brands, but through specialized platforms such as the Unmade technology company.

Despite the fact that, in addition to accelerating basic casteization processes, more complex tasks are being carried out in scientific laboratories, such as 2D-Lecal and 3D landing images based on [5] text, the more common option is not full automation, but the situation where AI acts as an assistant designer: offers options, checks errors, accelerates routine operations. Experimental studies emphasize the principle of augment, not replay, «add to, not replace» [6].

One notable trend in fashion in recent years has been the shift of focus from a silhouette to an invoice and material. Designers are increasingly working not so much with clothing, but rather with tissue surface, density, rhythm and tactile properties, linking complex invoices to graphics. Here, the use of AI is also required, especially in the creation of modern jackcard tissue. This approach is based on the so-called warp / webthinking method of design, where the image is thought not as a flat picture, but as a result of the intersection of the base (warp) and the duck (weft). Unlike simple automation, which mechanically translates the image into a binding scheme, AI allows the interpretation of visual material: to select key rhythms, contrast levels and structurally important elements, adapting them to weaving limitations and helping design the fabrics in which the image is embedded in the structure of the material. This is particularly important in dealing with complex graphics, printing, logos and multilayered images.

This approach is similar to the practice Concept Carpet Lab , a creative laboratory that works with a rug as a psychic of relevant art and rethinks the traditions of carpeting through modern authorship. The debut collection was created for Igor Gurovich’s work, and then the lab continued to work in a collaboration with Yuri Gordon, both of which do not act as decorative objects, but as independent artwork and as a form of visual storitelling.

Unlike generic tools, a competer view in factories is no longer an experiment, but a routine that helps detect tissue defects, control stitches, and test ready-made products for marriage. Studies in textiles have shown that such systems reduce the number of marriages and speed up the screening process without increasing the burden on workers [7].

The second part deals with areas where technology becomes directly visible to the audience: virtual fitting, digital-model-fashing, brand marketing language, and new requirements for transparency and regulation of AI content.

Sources: